Tiago Quelhas

Would it help to capture execution logs for local and remote builds and compare them? (See https://bazel.build/remote/cache-remote#caching-across-machines)

FYI, the changes to `remote_execution.proto` must be sent in a separate PR. I can do that tomorrow if you don't beat me to it.

> @tjgq I updated this to say "Fixes #8421" as it realizes inlining for the cases that were originally anticipated. I think it's better to open a dedicated issue for...

I will do the import and resolve the conflicts.

From offline discussion: https://github.com/bazelbuild/bazel/pull/23428 will mark in-memory outputs (currently: `.d`, `.jdeps` and `unused_inputs_list` files) as inlinable. Beyond that, we should consider marking any output that will be immediately downloaded (per...

That's a good point - forcing the download to occur synchronously might very well undo the gains of avoiding the extra round trip. I suppose we could try it out...

(Also note: while we *could* download outputs in the background, they currently block completion of the spawn that produces them: https://cs.opensource.google/bazel/bazel/+/master:src/main/java/com/google/devtools/build/lib/remote/RemoteExecutionService.java;l=1239;drc=d91e05e4a82e61720d41e64b81d4b36924b1c398).

Yeah, I'm now of the opinion that we should move downloads to the background (other than the in-memory case) and not bother with inlining them at all.

@coeuvre WDYT? In particular, is it worth it to put this behind a flag, or should we just do it unconditionally?

Thanks, I've sent https://github.com/bazelbuild/proposals/pull/328 to fix the example in the proposal. More importantly, we need better user-facing documentation on how to write a credential helper - note that proposals only...